RESEARCH ARTICLE
Food for Pollinators: Quantifying the Nectar and Pollen Resources of Urban Flower Meadows Damien M. Hicks1, Pierre Ouvrard1,2, Katherine C. R. Baldock3,4, Mathilde Baude3,5, Mark A. Goddard6¤, William E. Kunin6, Nadine Mitschunas3,7, Jane Memmott3,4, Helen Morse3, Maria Nikolitsi6, Lynne M. Osgathorpe3, Simon G. Potts7, Kirsty M. Robertson6, Anna V. Scott7, Frazer Sinclair1,8, Duncan B. Westbury9, Graham N. Stone1*
a11111
OPEN ACCESS
1 Institute of Evolutionary Biology, University of Edinburgh, Kings Buildings, Charlotte Auerbach Road, Edinburgh EH9 3JT, United Kingdom, 2 Earth and Life Institute - Agronomy, Université catholique de Louvain, Place Croix du Sud 2, 1348 Louvain-la-Neuve, Belgium, 3 School of Biological Sciences, University of Bristol, 24 Tyndall Avenue, Bristol, BS8 1TQUG, United Kingdom, 4 Cabot Institute, University of Bristol, Woodland Road, Bristol, BS8 1UJ, United Kingdom, 5 Collegium Sciences et Techniques, EA 1207 LBLGC, Université d’Orléans, 45067, Orléans, France, 6 School of Biology, University of Leeds, Leeds, LS2 9JT, United Kingdom, 7 Centre for Agri-Environmental Research, School of Agriculture, Policy and Development, University of Reading, Reading, RG6 6AR, United Kingdom, 8 Royal Society for the Protection of Birds, Gola Rainforest National Park, Kenema, Sierra Leone, 9 Institute of Science & the Environment, The University of Worcester, Henwick Grove, Worcester, WR2 6AJ, United Kingdom
Citation: Hicks DM, Ouvrard P, Baldock KCR, Baude M, Goddard MA, Kunin WE, et al. (2016) Food for Pollinators: Quantifying the Nectar and Pollen Resources of Urban Flower Meadows. PLoS ONE 11(6): e0158117. doi:10.1371/journal.pone.0158117
¤ Current address: School of Civil Engineering and Geosciences, Newcastle University, Newcastle upon Tyne, NE1 7RU, United Kingdom * [email protected]
Editor: Cheng–Sen Li, Institute of Botany, CHINA
Abstract
Received: November 20, 2015 Accepted: June 12, 2016 Published: June 24, 2016 Copyright: © 2016 Hicks et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All relevant data are provided within the paper and its Supporting Information files. Funding: This work was funded jointly by a grant from Biotechnology and Biological Sciences Research Council (BBSRC), The UK Government Department for Environment, Food and Rural Affairs (Defra), Natural Environment Research Council (NERC), the Scottish Government and the Wellcome Trust, under the Insect Pollinators Initiative (www. bbsrc.ac.uk/funding/opportunities/2009/insectpollinators-initiative/), (grant numbers BB/I000305/1, BB/I00047X/1) to Jane Memmott, Graham N. Stone, Simon G. Potts and William E. Kunin, a second Insect
Planted meadows are increasingly used to improve the biodiversity and aesthetic amenity value of urban areas. Although many ‘pollinator-friendly’ seed mixes are available, the floral resources these provide to flower-visiting insects, and how these change through time, are largely unknown. Such data are necessary to compare the resources provided by alternative meadow seed mixes to each other and to other flowering habitats. We used quantitative surveys of over 2 million flowers to estimate the nectar and pollen resources offered by two exemplar commercial seed mixes (one annual, one perennial) and associated weeds grown as 300m2 meadows across four UK cities, sampled at six time points between May and September 2013. Nectar sugar and pollen rewards per flower varied widely across 65 species surveyed, with native British weed species (including dandelion, Taraxacum agg.) contributing the top five nectar producers and two of the top ten pollen producers. Seed mix species yielding the highest rewards per flower included Leontodon hispidus, Centaurea cyanus and C. nigra for nectar, and Papaver rhoeas, Eschscholzia californica and Malva moschata for pollen. Perennial meadows produced up to 20x more nectar and up to 6x more pollen than annual meadows, which in turn produced far more than amenity grassland controls. Perennial meadows produced resources earlier in the year than annual meadows, but both seed mixes delivered very low resource levels early in the year and these were provided almost entirely by native weeds. Pollen volume per flower is well predicted statistically by floral morphology, and nectar sugar mass and pollen volume per unit area are correlated
PLOS ONE | DOI:10.1371/journal.pone.0158117 June 24, 2016
1 / 37
Quantifying Floral Resources in Urban Flower Meadows
Pollinators Initiative grant (grant number BB/ H014934/1) to William E. Kunin, Simon G. Potts and Jane Memmott, and Impact Acceleration Account funding from the Natural Environment Research Council to Graham N. Stone. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
with flower counts, raising the possibility that resource levels can be estimated for species or habitats where they cannot be measured directly. Our approach does not incorporate resource quality information (for example, pollen protein or essential amino acid content), but can easily do so when suitable data exist. Our approach should inform the design of new seed mixes to ensure continuity in floral resource availability throughout the year, and to identify suitable species to fill resource gaps in established mixes.
Competing Interests: The authors have declared that no competing interests exist.
Introduction There is increasing interest in the role of urban environments as habitats for wildlife, including species of conservation concern [1,2]. This interest stems from the fact that while natural habitats are declining and becoming increasingly fragmented in many parts of the world, urban habitats are expanding [2–6]. The current focus on pollinators is driven by the value of pollination services for supporting wider biodiversity and contributing to human food supplies (e.g. [7]), and growing evidence of diversity loss and range contractions for some species of wild pollinators in Europe and North America [8–10]. Studies to date (e.g. [11–17]) show that pollinator assemblages change along urbanisation gradients in a way that varies among pollinator taxa. While some urban environments support low pollinator diversity (e.g. [18]), others support high pollinator abundance and/or species richness [16,18–20]. A recent analysis of UK pollinator assemblages, for example, found cities to support higher bee species richness, but lower hoverfly abundance, than farmland [20]. Widespread public recognition of their ecological role has also made pollinators (and particularly bumblebees) an important flagship group for raising public awareness of human impacts on biodiversity [16]. Food limitation, resulting from decreasing flower diversity and quantity, is thought to be one of multiple causes for pollinator decline [21–24]. While dependence of pollinator populations on floral resources varies among taxa (see Discussion), the pollen harvested by bees is invested directly in the next generation [25–27]. Pollen from between 20 and several thousand flowers is required to rear a single solitary bee larva [22], and 120kg of nectar and 20kg of pollen are harvested annually by a single temperate European honeybee colony [28]. Growing numbers of studies have shown pollinator visitation to urban flowers [19,20,29–33] and a positive correlation between the abundances of flowers and pollinators [29,33–35]. These findings suggest that pollinator abundance and diversity could be increased by changes in urban land use that increase floral resource availability. Urban fruit and vegetable production is also probably dependent on local reservoirs of pollinators [30,36] whose life histories require sugar and protein at predictable points in the season. Recent years have seen increased planting of seed mixes in urban landscapes, creating extensive flowering borders and urban meadows (e.g. [37,38]). Such ‘green infrastructure’ mixes can comprise both native and exotic species, and are often designed to contain primarily herbs rather than grasses (in contrast to natural hay meadows) [39,40]. Selected sets of species are also chosen to provide a long season with a high intensity of flowering and hence aesthetic impact [38,41]. The motivation for planting urban meadows commonly combines benefits for human quality of life [42,43] with desire to increase the biodiversity value of urban spaces. This is actively encouraged by a range of civil and planning awards and initiatives, including the UK’s Britain in Bloom (https://www.rhs.org.uk/communities/campaigns/britain-in-bloom/) and Green Flag Awards (http://www.greenflagaward.org/), and Belgium’s Plan Maya (http:// biodiversite.wallonie.be/fr/plan-maya.html?IDC=5617). The expectation is that such flower-
PLOS ONE | DOI:10.1371/journal.pone.0158117 June 24, 2016
2 / 37
Quantifying Floral Resources in Urban Flower Meadows
rich patches will provide more nectar and pollen resources for pollinators than the frequently mowed amenity grassland (i.e. cultivated open park grassland) that makes up many urban green spaces. There is evidence that pollinators visit urban flower plantings more than unplanted comparison plots [44], paralleling observations for flower margins planted adjacent to agricultural crops [45–48]. Numerous commercial annual and perennial seed mixes are available for establishing flower-rich habitat (see S5 Table). However, we do not know the nectar and pollen resources per flower provided by different seed mix species, or resource levels per unit area of different seed mixes. These are important knowledge gaps, because seed merchants cannot currently design (or landscape managers select) seed mixes on the basis of their ability to provide resources to pollinators. Seasonal timing (phenology), quality and quantity of floral resources are all important for pollinator populations [25,26,49]. In particular, seed mixes designed to support pollinators must deliver pollen and nectar throughout the season, without dips in seasonal resource availability that could potentially limit pollinator populations [50]. Given the wide range of available seed mixes, here we focus on quantifying the per-species nectar and pollen resources in two commercially available exemplars, one annual and one perennial, which show substantial overlap in species composition with available alternatives (S5 Table). The annual “Rainbow” mix supplied by Rigby Taylor is a market leader in the UK that contains 14 native and non-native annual species, and in 2014 and 2015 was planted over more than 100 hectares (ha) of urban meadows. The perennial “Special Pollen and Nectar Wildflower” mix sold by Emorsgate Seeds contains 23 native species and over the last six years has been sown on over 30 ha of parks and gardens and (in a slightly modified form), on ca. 120 ha of farmland, primarily for UK agri-environment schemes. The two mixes in our study include all of the species most commonly included in a panel of 10 commercially available pollinator seed mixes (S5 Table): Centaurea cyanus, Leucanthemum vulgare (5 mixes), Centaurea nigra, Daucus carota, Lotus corniculatus, Silene dioica, and Trifolium pratense (4 mixes), making them appropriate exemplars for study. We planted each mix in replicate 300 m2 meadows in four UK towns and cities (Bristol, Edinburgh, Reading and Leeds) and compared the rewards provided by these meadows with control plots of amenity grassland. Because weed species such as Taraxacum agg. can provide significant resources to pollinators [51], we included non-mix (weed) species in surveys of all treatments. In this context, the weeds in our experiments comprise both native species and non-native garden-escape species, a common situation in urban environments. Our objectives are to (i) quantify the per-flower nectar and pollen resources provided by each seed mix and weed species, highlighting those providing high and low rewards; (ii) assess between-city variation in the composition of meadows resulting from planting the same mixes with the same protocols across the UK; and (iii) quantify meadow-level changes in floral resource provision through time and between meadow treatments. We explore the consequences of alternative sampling intensities for surveys of floral abundance and develop predictive statistical models for pollen volume per flower (based on floral morphology) and nectar and pollen resource per unit area (based on flower counts). Our approach does not incorporate variation in resource quality [52–54], but can easily be modified to do so where suitable data exist (see Discussion).
Materials and Methods Seed mixes and experimental design We used two commercially available flower seed mixes, one annual and the other perennial: Rigby Taylor’s ‘Rainbow Annual’ mix (14 species), and Emorsgate’s ‘EN1F Special Pollen and Nectar Wildflowers’ mix (23 species). Species compositions for these mixes, native/exotic
PLOS ONE | DOI:10.1371/journal.pone.0158117 June 24, 2016
3 / 37
Quantifying Floral Resources in Urban Flower Meadows
status [55], and common names for all mix and weed species are given in S1 Table. These two mix types characterise the main alternatives available to landscape managers. Annual seed mixes provide a quick return by flowering in their first year, but they generally need to be replanted each year to maintain performance. Disturbance of the ground in the first year of planting a seed mix has the potential to trigger weed growth from the seed bank, which could result in increasing weed growth in meadows grown on the same site as bare soil is repeatedly exposed in subsequent years. We assess the significance of this effect here by comparing resource provision in annual meadows in their first year, and in the same seed mix replanted on the same site in a second year. Commercially available annual mixes often contain a mixture of native, naturalised and non-native species: examples in the mix we used include Malcolmia maritima native to Mediterranean Europe, and multiple American species (Coreopsis species, Eschscholzia californica, Thelesperma burridgeanum and Cosmos bipinnatus), included to provide a showy display throughout a long flowering season. In contrast, perennial seed mixes take longer to become established and can require additional labour costs for weeding in the year before they flower [38], but need replanting much less often. The seeds in the perennial mix we used were all sourced from native UK provenances. Our experimental design comprised four treatments applied to amenity grassland in 80 sites across four UK towns and cities (Edinburgh, Leeds, Bristol, Reading, all termed cities hereafter), each having a total of 20 sites. In each city, we collected data in 2013 for five 300m2 replicates of each of the following four treatments: (i) annual mix sown in 2013 and sampled in its first year (A1); (ii) annual mix sown in 2012 and sampled after reseeding in 2013 in the same location (A2); (iii) perennial mix sown in 2012 and sampled in its second year in 2013, (iv) unplanted amenity grassland control, mown approximately every two weeks through the season, also sampled in 2013. Each treatment replicate had an area of 300 m2, with meadow shape varying among replicates as circumstances dictated. In each city, floral abundance in each treatment was surveyed at six time points between early May and early September 2013. Locations for all treatment replicates are detailed in S2 Table. To facilitate establishment of planted species and to maintain public acceptance of planted meadows, the largest and most visible weeds were removed at intervals. This included taxa (such as Sonchus species) subsequently shown to be highly rewarding, such that our resource estimates for weeds underestimate potential contributions from unweeded meadows. Large weed removal is nevertheless widely practised in urban green spaces, and our results are appropriate for such management.
Quantification of floral resource per flower We quantified floral resources per species in terms of daily nectar sugar mass and pollen volume [56] using protocols matching [57] for nectar and similar to those used in other studies [22,58]. We outline sampling methodologies below and provide detailed protocols in S2 and S3 Files. Floral resource measurements were made at the level of single flowers for all taxa except Asteraceae, for which resources were sampled at the level of the capitulum. For simplicity we use ‘flowers’ hereafter, and highlight contrasts between flowers and capitula where necessary. Grasses, sedges and wind-pollinated forbs were not sampled. Floral resource sampling for all species was carried out in Edinburgh in 2012 and 2013, using plants sampled as available across all 15 planted and five control replicate meadows. We recognise that our sampling thus does not incorporate environmental impacts on quantity or quality of resource provision by seeds of the same cultivars grown in different cities. To assess the extent to which our resource values agree with other estimates for the same species, we correlated our nectar resource data with values generated using the same methods recently published by Baude et al. [57].
PLOS ONE | DOI:10.1371/journal.pone.0158117 June 24, 2016
4 / 37
Quantifying Floral Resources in Urban Flower Meadows
(i) Nectar sampling. On days with no rain, nectar was allowed to accumulate for 24 hours in flowers from which insects were excluded by fine netting. Nectar quantification followed one of two protocols (see S2 File for detailed methods). When possible, nectar was collected directly using 1.0 μl microcapillaries (VWR International, UK) and the degrees Brix sugar content (g of sugar in 100g solution) estimated using a sucrose refractometer (VWR International, UK, using refractometers modified by the maker to accept very low volumes). Nectar sugar content of each sample (mg) was calculated from the equation s = 10dvC, where v is the volume of nectar (ml), and d is the density of a sucrose solution at a concentration C (g sucrose/100 g solution) as read on the refractometer [59,60]. The density was obtained as d = 0.0037921C + 0.0000178C2 + 0.9988603 [60]. We sampled 9.5 ± 0.6 (range 3–22) flowers per species (depending on availability) to estimate a species mean (see S1 Table). Where volumes of nectar were too small or concentrations of nectar too high to use this approach, nectar sugar was harvested by using a fine Gilson pipette to flush flowers with a known volume of distilled water and the Brix sugar content of the resulting solution estimated using a sucrose refractometer [57]. The mass of sugar contained was estimated as above. Nectar reward values are presented as mean sugar mass/floral unit ± 1 standard error of the mean. Nectar analyses covered 80% of species and over 99% of all flowers recorded in our surveys. (ii) Pollen sampling. Pollen per flower was estimated for all species using flowers collected in Edinburgh treatment meadows in 2012 and 2013, for a total of 15,925 flowers across 64 species (see S3 File for detailed methods). Pollen was harvested using sonication (Dawe sonicleaner) from flowers that opened and dehisced in the lab. The collected pollen was suspended in 70% ethanol, and the number of pollen grains in a known volume aliquot counted on a haemocytometer slide. We estimated the volume of a sample of 100 pollen grains of each species using the formula for an ellipsoid (Volume = (4/3) π (A/2) (B/2)2, where ‘a’ is the major axis and ‘b’ the minor axis of the pollen grain). Pollen volume per floral unit was then calculated as (total number of pollen grains/floral unit) x (mean volume per pollen grain). Sample size information for resource quantification in each species is provided in S1 Table, from a mean of 10.3 ± 0.6 (range 2–22) floral units/species, depending on availability. Pollen resource per floral unit per 24h was estimated by dividing the pollen volume per floral unit by floral longevity in days. This approach does not assume that pollen release is constant during the lifetime of each flower, but does assume that the proportion of flowers in the sampled population in each stage of pollen release (if it varies through floral life) was stable at the time longevity was measured (see below). Pollen resource values are presented as mean pollen volume/floral unit ± 1 standard error of the mean. Pollen resource per 24h was estimated for 78% of the species recorded in treatment plots, representing over 99% of the flowers present in our surveys (see S1 Table). We explored the predictability of pollen volume per flower (or per floret for Asteraceae) from floral morphology, using a linear model incorporating anther size and number of stamens as predictors in the base R package [61]. To simplify use of the model, we allocated anther sizes for each species to one of four size classes (S3 Table): ‘tiny’ (for all Asteraceae), ‘small’ (e.g. Galium verum, Stellaria media; to a maximum anther volume of ca. 1.25mm3), ‘medium’ (e.g. Ranunculus spp., Malcolmia maritima; ca. 1.25–2.25mm3) and ‘large’ (e.g. Papaver rhoeas, Chamerion angustifolium: volume > 2.25mm3). Data on stamen numbers/floral unit are also provided in S3 Table. Normalising transformations for the parameters (loge pollen per flower and 1/stamen number) were estimated by likelihood ratio tests using the powerTransform function in {car} [62], which maps variables to a family of functions to create a rank-preserving transformation of data using power functions. The regression model was fitted and checked for non-violation of assumptions of linearity, normality and homoscedasticity by inspection of residuals, fitted values and quantile plots.
PLOS ONE | DOI:10.1371/journal.pone.0158117 June 24, 2016
5 / 37
Quantifying Floral Resources in Urban Flower Meadows
We tested the predictive power of our model by estimating per-flower pollen volumes for a validation data set comprising 43 species quantified using the same lab protocol (Baude et al. unpublished data; see [57] for information on the Agriland Project generating these data). Species in the validation data set were selected to span the same range of measured pollen volumes. We attempted to develop similar predictive models for nectar sugar per flower based on the floral morphology traits of corolla diameter and corolla length. No useful predictive model resulted from this approach. (iii) Floral longevity. Floral unit longevity was estimated by scoring the numbers of newly opening and newly closed (i.e. post-reproductive) flowers entering a fixed survey area (see S3 File for detailed methodology). Under the assumption that the population of flowers of a given species is approximately stable, floral longevity in days can be estimated as (2a+(b-c)) / ((b-c) d), where a = total open flowers on first observation, b = total newly-open flowers on second observation, c = total newly-closed flowers on second observation, and d = number of days between first and second observation. To meet the assumptions of this approach as closely as possible, species were observed in the middle of their flowering period. To minimise weather impacts on resources, observations were set up when the weather forecast was for no rain for the following day and the temperature was above 15°C. Sampling included the full range of flower ages, totalling approximately 50 flowers distributed on at least five plants. After marking the boundary of the study area and/or the inflorescences to be included, the area was surveyed on two occasions exactly 24 hours apart (or a multiple thereof for longer-lasting flowers). Longevity data for almost all species were collected for plants from Edinburgh meadows, with additional data where necessary from sites specified in S1 Table, sampled using the same protocol.
Surveys of floral abundance and resources per meadow We used quadrat surveys to quantify floral abundance in all meadows, at the level of individual flowers for all species except Asteraceae, which were sampled at the level of the capitulum. To identify an appropriate sampling scheme for meadow-wide resources, we conducted pilot surveys of A2 meadows in their first year at three Edinburgh sites in 2012 (Inch, Cairntows, Davidson’s Mains). We sampled between 49 and 99 quadrats for each meadow, and then used computer simulations (sampling actual transect values at random, without replacement) to assess the impact of sampling increasing numbers of 1m2 quadrats on the cumulative mean of nectar sugar mass and pollen volume per meadow (S1 and S2 Figs). In each meadow, variance in mean estimates declined sharply with increasing sample size, stabilising for each meadow and resource at a sample size of ca. 20 quadrats. We therefore sampled 20 x 1m2 quadrats per treatment replicate per sampling interval for the Edinburgh meadows during the main 2013 survey season (a total of 1400 quadrats over all treatments and sampling intervals). The 20 quadrats were sampled at 4 m intervals along the length of the plot, alternating between edge quadrats (i.e. extending from the plot edge to 1m into the plot) and internal quadrats (i.e. extending from 1m to 2m into the plot). For the three other cities, time constraints restricted sampling to a subset of the 20-quadrat scheme, comprising seven edge quadrats per treatment replicate per sampling interval located at 8m intervals along the plot length. We compare the consequences of adopting the seven or 20-quadrat scheme below using the Edinburgh data. In short, we consider the seven-quadrat estimates adequate for comparison of mean differences between treatments, and all comparisons among cities are based on data for the seven-quadrat scheme. Total numbers of flowers (or capitula) per treatment replicate for each species were estimated as the product of mean floral density/m2 x meadow area (m2). In 2013, over the four cities, 80 plots and six sampling intervals, we recorded over two million flowers of 105 plant species, from a total of 3068 quadrats.
PLOS ONE | DOI:10.1371/journal.pone.0158117 June 24, 2016
6 / 37
Quantifying Floral Resources in Urban Flower Meadows
To illustrate variation among replicates we present detailed data summaries for each of the Edinburgh treatment replicates, and present the data for each meadow location in the other three cities in S4 and S5 Figs. To allow comparison between cities, treatments and sampling time points, we use means calculated for each treatment across all time points, replicates and cities, as appropriate. We excluded from these values data for four sites at which management issues (detailed below) led to complete failure of the treatment relative to others in the same group. These comprise two perennial sites and two A1 annual sites (S2 Table). Resources per meadow (nectar sugar mass or pollen volume) were calculated as the sum across plant species of (resource per flower) x (the number of flowers in the treatment replicate). 95% confidence intervals for resources are based on error estimates in the floral counts.
Statistical methods Statistical analyses were carried out in R version 3.1.3 (R Foundation for Statistical Computing, Vienna, Austria. 2015). To visualise seasonal trends in resource availability, means and 95% confidence intervals were plotted with the {ggplot2} package [63]. Correlations between each of nectar sugar and pollen volume and flower counts per transect were plotted with {PerformanceAnalytics} [64], and ranked dotplots were plotted with {stats} [65]. Flower counts (per species and summed across species) and resource values showed non-normality and heteroscedasticity that could not be corrected using standard data transformations. Variation of floral resource estimates among treatments was therefore tested using global Kruskal-Wallis tests on means calculated across all surveys for a given type (i.e. n = 7 quadrats x 5 replicates/city x 4 cities x 6 survey rounds) in {PMCMR} [66]. Pairwise comparisons between treatment means were tested using the post-hoc Tukey-Kramer-Nemenyi test with Chi-square correction for ties in {PMCMR}. Variation in the composition of planted floral meadows within a meadow treatment between cities and sampling points was tested using the manyglm function of the R package {mvabund} [67]. This approach allows identification of multivariate patterns and fitting of a separate Generalised Linear Model (GLM) for each flowering species using a common set of explanatory variables. Resampling-based hypothesis testing within mvabund was then used to make community-level and taxon-specific inferences about which factors were associated with the multivariate abundance patterns. For these analyses we used mean floral counts/m2 from seven quadrats for each treatment replicate (Annual A1, Annual A2, Perennial) per survey round (May, June, July, early August, late August, September) per city (Bristol, Edinburgh, Leeds, Reading). For this dataset we specified a negative binomial error distribution, and checked assumptions of mean-variance and log-linearity as detailed in [67], both by plotting directly and by plotting residuals versus fitted values. We used Monte Carlo bootstrapping to estimate p-values adjusted to control the family-wise error rate across species, at the default setting of 1000 resamples. Variation in floral composition was visualised using non-metric multidimensional scaling (NMDS) with the metaMDS function of {vegan} [68], using a Bray-Curtis distance matrix calculated using mean floral counts (capitula for Asteraceae) per species per m2 for each meadow location and sampling time point. Meadows with fewer than 10 flowers (or capitula) in total per m2 were excluded from the analysis. Because NMDS only uses rank information and maps ranks non-linearly onto ordination space, it can handle non-linear species responses of any shape and robustly find underlying gradients. As the final ordination is partly dependent on the initial configuration, we used up to 99 consecutive NMDS iterations with random starting configurations to test for stability of the result. All raw data, metadata and R code are available from the authors on request.
PLOS ONE | DOI:10.1371/journal.pone.0158117 June 24, 2016
7 / 37
Quantifying Floral Resources in Urban Flower Meadows
Results Resource per flower (a) Nectar sugar. Nectar sugar estimates per 24h for the seed mix species and associated weeds are shown in Fig 1 and S1 Table. Most of the highest values were for Asteraceae, for which a floral unit comprises an inflorescence (capitulum) rather than a single flower (Fig 1). The top-ranked annual species/floral unit/24h were Centaurea cyanus (896 ± 174μg s.e.m.), Cosmos bipinnatus (701 ± 69μg) and Calendula officinalis (470 ± 11μg). The top-ranked perennial species were Leontodon hispidus (1827 ± 193μg), Centaurea nigra (1474 ± 76μg) and Echium vulgare (688 ± 103μg). The top-ranked weed species, all native Asteraceae, produced more nectar sugar per floral unit than any seed mix species: Senecio jacobaea (2921 ± 448μg), Cirsium arvense (2609 +/- 239μg), Cirsium vulgare (2323 ± 418μg), and Taraxacum agg. (2137 ± 286μg). Both the annual and perennial mixes contained species with very low nectar sugar rewards per floral unit, including species with large individual flowers such as Papaver rhoeas (0.6 ± 0.6μg) and Eschscholzia californica (10 ± 1μg) in the annual mix, and species sampled as small individual flowers such as Lobularia maritima (4 ± 1μg) in the annual mix and Daucus carota (27 ± 7μg) and Galium verum (3.2 ± 0.6μg) in the perennial mix.
Fig 1. Mean nectar sugar mass per 24h per floral unit for species in A. the annual seed mix, B. the perennial seed mix, and B. native weeds in either mix. Values shown are ranked means in each group (mean values and standard errors are provided in S1 Table). Images of the top ranked species in each group are shown, with the highest-ranked at right. Images are provided by the project team with the exceptions of Echium vulgare (author: Ewan Cole) provided in 2016 under a (CCAL) CC BY 4.0 license from the Urban Flora of Scotland project. doi:10.1371/journal.pone.0158117.g001
PLOS ONE | DOI:10.1371/journal.pone.0158117 June 24, 2016
8 / 37
Quantifying Floral Resources in Urban Flower Meadows
(b) Pollen. Total pollen rewards per floral unit were highest in Asteraceae sampled at the level of the entire capitulum (Leucanthemum vulgare, 15.9 ± 2μl, Cosmos bipinnatus 13.8 ± 1.9μl), and lowest in those species sampled at the level of small individual flowers (e.g. Myosotis arvensis, 0.0004 ± 0.0001μl) (S1 Table). Of species sampled at the level of a single flower, by far the most rewarding were the poppies, Papaver rhoeas (13.3 ± 2.8μl) and Eschscholzia californica (8.3±1.1μl). Among weed species, values were again highest for Asteraceae, including Glebionis segetum (5.1 ± 0.9μl) (presumed to be a garden escape) and native Taraxacum agg. (2.8 ± 0.7μl). Quantifying the contribution of each species to daily pollen resource provision at the meadow level requires scaling of total pollen volume by floral longevity, which ranged from mean values of a single day (e.g. Cerastium fontanum, Veronica persica, Vicia hirsuta) to 14.8 days (Leucanthemum vulgare), and was generally higher in Asteraceae species sampled at the level of the capitulum (see Fig 2 and S1 Table). The highest pollen rewards/floral unit/24h were provided by annual mix species; flowers of Papaver rhoeas (6.0μl pollen) provided more than twice the values for the next-ranked species, Eschscholzia californica (2.4μl) and Calendula officinalis (1.8μl; this species ranked highly for both nectar and pollen rewards). In the annual mix, Coreopsis picta (0.7μl) had a very low pollen reward per capitulum in comparison to other
Fig 2. Mean pollen volume per 24h per floral unit for species in A. the annual seed mix, B. the perennial seed mix, and C. native weeds in either mix. Values shown are ranked means in each group. Mean values and standard errors, longevity and pollen volume/floral unit are provided in S1 Table). Images of the top ranked species in each group are shown, with the highest-ranked at right. Images are provided by the project team with the exceptions of Chamerion angustifolium (author: Ewan Cole) provided in 2016 under a (CCAL) CC BY 4.0 license by the Urban Flora of Scotland project. doi:10.1371/journal.pone.0158117.g002
PLOS ONE | DOI:10.1371/journal.pone.0158117 June 24, 2016
9 / 37
Quantifying Floral Resources in Urban Flower Meadows
Asteraceae sampled. The top-ranked perennial mix species by floral unit were Malva moschata (2.3μl), Centaurea nigra (2.1μl; this species ranked highly for both nectar and pollen rewards) and Leucanthemum vulgare (1.1μl). The top-ranked weed species in our meadows were native Taraxacum agg. (1.25μl) and Chamerion angustifolium (0.7μl). (c) Statistical prediction of resource per flower from floral morphology. Laboratory estimates of total pollen volume per floral unit were predicted by a linear model using anther size and stamen number as predictors (S3 Fig; adjusted R2 = 0.7025, F4,54 = 35.23, p
Food for Pollinators: Quantifying the Nectar and Pollen Resources of Urban Flower Meadows Damien M. Hicks1, Pierre Ouvrard1,2, Katherine C. R. Baldock3,4, Mathilde Baude3,5, Mark A. Goddard6¤, William E. Kunin6, Nadine Mitschunas3,7, Jane Memmott3,4, Helen Morse3, Maria Nikolitsi6, Lynne M. Osgathorpe3, Simon G. Potts7, Kirsty M. Robertson6, Anna V. Scott7, Frazer Sinclair1,8, Duncan B. Westbury9, Graham N. Stone1*
a11111
OPEN ACCESS
1 Institute of Evolutionary Biology, University of Edinburgh, Kings Buildings, Charlotte Auerbach Road, Edinburgh EH9 3JT, United Kingdom, 2 Earth and Life Institute - Agronomy, Université catholique de Louvain, Place Croix du Sud 2, 1348 Louvain-la-Neuve, Belgium, 3 School of Biological Sciences, University of Bristol, 24 Tyndall Avenue, Bristol, BS8 1TQUG, United Kingdom, 4 Cabot Institute, University of Bristol, Woodland Road, Bristol, BS8 1UJ, United Kingdom, 5 Collegium Sciences et Techniques, EA 1207 LBLGC, Université d’Orléans, 45067, Orléans, France, 6 School of Biology, University of Leeds, Leeds, LS2 9JT, United Kingdom, 7 Centre for Agri-Environmental Research, School of Agriculture, Policy and Development, University of Reading, Reading, RG6 6AR, United Kingdom, 8 Royal Society for the Protection of Birds, Gola Rainforest National Park, Kenema, Sierra Leone, 9 Institute of Science & the Environment, The University of Worcester, Henwick Grove, Worcester, WR2 6AJ, United Kingdom
Citation: Hicks DM, Ouvrard P, Baldock KCR, Baude M, Goddard MA, Kunin WE, et al. (2016) Food for Pollinators: Quantifying the Nectar and Pollen Resources of Urban Flower Meadows. PLoS ONE 11(6): e0158117. doi:10.1371/journal.pone.0158117
¤ Current address: School of Civil Engineering and Geosciences, Newcastle University, Newcastle upon Tyne, NE1 7RU, United Kingdom * [email protected]
Editor: Cheng–Sen Li, Institute of Botany, CHINA
Abstract
Received: November 20, 2015 Accepted: June 12, 2016 Published: June 24, 2016 Copyright: © 2016 Hicks et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All relevant data are provided within the paper and its Supporting Information files. Funding: This work was funded jointly by a grant from Biotechnology and Biological Sciences Research Council (BBSRC), The UK Government Department for Environment, Food and Rural Affairs (Defra), Natural Environment Research Council (NERC), the Scottish Government and the Wellcome Trust, under the Insect Pollinators Initiative (www. bbsrc.ac.uk/funding/opportunities/2009/insectpollinators-initiative/), (grant numbers BB/I000305/1, BB/I00047X/1) to Jane Memmott, Graham N. Stone, Simon G. Potts and William E. Kunin, a second Insect
Planted meadows are increasingly used to improve the biodiversity and aesthetic amenity value of urban areas. Although many ‘pollinator-friendly’ seed mixes are available, the floral resources these provide to flower-visiting insects, and how these change through time, are largely unknown. Such data are necessary to compare the resources provided by alternative meadow seed mixes to each other and to other flowering habitats. We used quantitative surveys of over 2 million flowers to estimate the nectar and pollen resources offered by two exemplar commercial seed mixes (one annual, one perennial) and associated weeds grown as 300m2 meadows across four UK cities, sampled at six time points between May and September 2013. Nectar sugar and pollen rewards per flower varied widely across 65 species surveyed, with native British weed species (including dandelion, Taraxacum agg.) contributing the top five nectar producers and two of the top ten pollen producers. Seed mix species yielding the highest rewards per flower included Leontodon hispidus, Centaurea cyanus and C. nigra for nectar, and Papaver rhoeas, Eschscholzia californica and Malva moschata for pollen. Perennial meadows produced up to 20x more nectar and up to 6x more pollen than annual meadows, which in turn produced far more than amenity grassland controls. Perennial meadows produced resources earlier in the year than annual meadows, but both seed mixes delivered very low resource levels early in the year and these were provided almost entirely by native weeds. Pollen volume per flower is well predicted statistically by floral morphology, and nectar sugar mass and pollen volume per unit area are correlated
PLOS ONE | DOI:10.1371/journal.pone.0158117 June 24, 2016
1 / 37
Quantifying Floral Resources in Urban Flower Meadows
Pollinators Initiative grant (grant number BB/ H014934/1) to William E. Kunin, Simon G. Potts and Jane Memmott, and Impact Acceleration Account funding from the Natural Environment Research Council to Graham N. Stone. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
with flower counts, raising the possibility that resource levels can be estimated for species or habitats where they cannot be measured directly. Our approach does not incorporate resource quality information (for example, pollen protein or essential amino acid content), but can easily do so when suitable data exist. Our approach should inform the design of new seed mixes to ensure continuity in floral resource availability throughout the year, and to identify suitable species to fill resource gaps in established mixes.
Competing Interests: The authors have declared that no competing interests exist.
Introduction There is increasing interest in the role of urban environments as habitats for wildlife, including species of conservation concern [1,2]. This interest stems from the fact that while natural habitats are declining and becoming increasingly fragmented in many parts of the world, urban habitats are expanding [2–6]. The current focus on pollinators is driven by the value of pollination services for supporting wider biodiversity and contributing to human food supplies (e.g. [7]), and growing evidence of diversity loss and range contractions for some species of wild pollinators in Europe and North America [8–10]. Studies to date (e.g. [11–17]) show that pollinator assemblages change along urbanisation gradients in a way that varies among pollinator taxa. While some urban environments support low pollinator diversity (e.g. [18]), others support high pollinator abundance and/or species richness [16,18–20]. A recent analysis of UK pollinator assemblages, for example, found cities to support higher bee species richness, but lower hoverfly abundance, than farmland [20]. Widespread public recognition of their ecological role has also made pollinators (and particularly bumblebees) an important flagship group for raising public awareness of human impacts on biodiversity [16]. Food limitation, resulting from decreasing flower diversity and quantity, is thought to be one of multiple causes for pollinator decline [21–24]. While dependence of pollinator populations on floral resources varies among taxa (see Discussion), the pollen harvested by bees is invested directly in the next generation [25–27]. Pollen from between 20 and several thousand flowers is required to rear a single solitary bee larva [22], and 120kg of nectar and 20kg of pollen are harvested annually by a single temperate European honeybee colony [28]. Growing numbers of studies have shown pollinator visitation to urban flowers [19,20,29–33] and a positive correlation between the abundances of flowers and pollinators [29,33–35]. These findings suggest that pollinator abundance and diversity could be increased by changes in urban land use that increase floral resource availability. Urban fruit and vegetable production is also probably dependent on local reservoirs of pollinators [30,36] whose life histories require sugar and protein at predictable points in the season. Recent years have seen increased planting of seed mixes in urban landscapes, creating extensive flowering borders and urban meadows (e.g. [37,38]). Such ‘green infrastructure’ mixes can comprise both native and exotic species, and are often designed to contain primarily herbs rather than grasses (in contrast to natural hay meadows) [39,40]. Selected sets of species are also chosen to provide a long season with a high intensity of flowering and hence aesthetic impact [38,41]. The motivation for planting urban meadows commonly combines benefits for human quality of life [42,43] with desire to increase the biodiversity value of urban spaces. This is actively encouraged by a range of civil and planning awards and initiatives, including the UK’s Britain in Bloom (https://www.rhs.org.uk/communities/campaigns/britain-in-bloom/) and Green Flag Awards (http://www.greenflagaward.org/), and Belgium’s Plan Maya (http:// biodiversite.wallonie.be/fr/plan-maya.html?IDC=5617). The expectation is that such flower-
PLOS ONE | DOI:10.1371/journal.pone.0158117 June 24, 2016
2 / 37
Quantifying Floral Resources in Urban Flower Meadows
rich patches will provide more nectar and pollen resources for pollinators than the frequently mowed amenity grassland (i.e. cultivated open park grassland) that makes up many urban green spaces. There is evidence that pollinators visit urban flower plantings more than unplanted comparison plots [44], paralleling observations for flower margins planted adjacent to agricultural crops [45–48]. Numerous commercial annual and perennial seed mixes are available for establishing flower-rich habitat (see S5 Table). However, we do not know the nectar and pollen resources per flower provided by different seed mix species, or resource levels per unit area of different seed mixes. These are important knowledge gaps, because seed merchants cannot currently design (or landscape managers select) seed mixes on the basis of their ability to provide resources to pollinators. Seasonal timing (phenology), quality and quantity of floral resources are all important for pollinator populations [25,26,49]. In particular, seed mixes designed to support pollinators must deliver pollen and nectar throughout the season, without dips in seasonal resource availability that could potentially limit pollinator populations [50]. Given the wide range of available seed mixes, here we focus on quantifying the per-species nectar and pollen resources in two commercially available exemplars, one annual and one perennial, which show substantial overlap in species composition with available alternatives (S5 Table). The annual “Rainbow” mix supplied by Rigby Taylor is a market leader in the UK that contains 14 native and non-native annual species, and in 2014 and 2015 was planted over more than 100 hectares (ha) of urban meadows. The perennial “Special Pollen and Nectar Wildflower” mix sold by Emorsgate Seeds contains 23 native species and over the last six years has been sown on over 30 ha of parks and gardens and (in a slightly modified form), on ca. 120 ha of farmland, primarily for UK agri-environment schemes. The two mixes in our study include all of the species most commonly included in a panel of 10 commercially available pollinator seed mixes (S5 Table): Centaurea cyanus, Leucanthemum vulgare (5 mixes), Centaurea nigra, Daucus carota, Lotus corniculatus, Silene dioica, and Trifolium pratense (4 mixes), making them appropriate exemplars for study. We planted each mix in replicate 300 m2 meadows in four UK towns and cities (Bristol, Edinburgh, Reading and Leeds) and compared the rewards provided by these meadows with control plots of amenity grassland. Because weed species such as Taraxacum agg. can provide significant resources to pollinators [51], we included non-mix (weed) species in surveys of all treatments. In this context, the weeds in our experiments comprise both native species and non-native garden-escape species, a common situation in urban environments. Our objectives are to (i) quantify the per-flower nectar and pollen resources provided by each seed mix and weed species, highlighting those providing high and low rewards; (ii) assess between-city variation in the composition of meadows resulting from planting the same mixes with the same protocols across the UK; and (iii) quantify meadow-level changes in floral resource provision through time and between meadow treatments. We explore the consequences of alternative sampling intensities for surveys of floral abundance and develop predictive statistical models for pollen volume per flower (based on floral morphology) and nectar and pollen resource per unit area (based on flower counts). Our approach does not incorporate variation in resource quality [52–54], but can easily be modified to do so where suitable data exist (see Discussion).
Materials and Methods Seed mixes and experimental design We used two commercially available flower seed mixes, one annual and the other perennial: Rigby Taylor’s ‘Rainbow Annual’ mix (14 species), and Emorsgate’s ‘EN1F Special Pollen and Nectar Wildflowers’ mix (23 species). Species compositions for these mixes, native/exotic
PLOS ONE | DOI:10.1371/journal.pone.0158117 June 24, 2016
3 / 37
Quantifying Floral Resources in Urban Flower Meadows
status [55], and common names for all mix and weed species are given in S1 Table. These two mix types characterise the main alternatives available to landscape managers. Annual seed mixes provide a quick return by flowering in their first year, but they generally need to be replanted each year to maintain performance. Disturbance of the ground in the first year of planting a seed mix has the potential to trigger weed growth from the seed bank, which could result in increasing weed growth in meadows grown on the same site as bare soil is repeatedly exposed in subsequent years. We assess the significance of this effect here by comparing resource provision in annual meadows in their first year, and in the same seed mix replanted on the same site in a second year. Commercially available annual mixes often contain a mixture of native, naturalised and non-native species: examples in the mix we used include Malcolmia maritima native to Mediterranean Europe, and multiple American species (Coreopsis species, Eschscholzia californica, Thelesperma burridgeanum and Cosmos bipinnatus), included to provide a showy display throughout a long flowering season. In contrast, perennial seed mixes take longer to become established and can require additional labour costs for weeding in the year before they flower [38], but need replanting much less often. The seeds in the perennial mix we used were all sourced from native UK provenances. Our experimental design comprised four treatments applied to amenity grassland in 80 sites across four UK towns and cities (Edinburgh, Leeds, Bristol, Reading, all termed cities hereafter), each having a total of 20 sites. In each city, we collected data in 2013 for five 300m2 replicates of each of the following four treatments: (i) annual mix sown in 2013 and sampled in its first year (A1); (ii) annual mix sown in 2012 and sampled after reseeding in 2013 in the same location (A2); (iii) perennial mix sown in 2012 and sampled in its second year in 2013, (iv) unplanted amenity grassland control, mown approximately every two weeks through the season, also sampled in 2013. Each treatment replicate had an area of 300 m2, with meadow shape varying among replicates as circumstances dictated. In each city, floral abundance in each treatment was surveyed at six time points between early May and early September 2013. Locations for all treatment replicates are detailed in S2 Table. To facilitate establishment of planted species and to maintain public acceptance of planted meadows, the largest and most visible weeds were removed at intervals. This included taxa (such as Sonchus species) subsequently shown to be highly rewarding, such that our resource estimates for weeds underestimate potential contributions from unweeded meadows. Large weed removal is nevertheless widely practised in urban green spaces, and our results are appropriate for such management.
Quantification of floral resource per flower We quantified floral resources per species in terms of daily nectar sugar mass and pollen volume [56] using protocols matching [57] for nectar and similar to those used in other studies [22,58]. We outline sampling methodologies below and provide detailed protocols in S2 and S3 Files. Floral resource measurements were made at the level of single flowers for all taxa except Asteraceae, for which resources were sampled at the level of the capitulum. For simplicity we use ‘flowers’ hereafter, and highlight contrasts between flowers and capitula where necessary. Grasses, sedges and wind-pollinated forbs were not sampled. Floral resource sampling for all species was carried out in Edinburgh in 2012 and 2013, using plants sampled as available across all 15 planted and five control replicate meadows. We recognise that our sampling thus does not incorporate environmental impacts on quantity or quality of resource provision by seeds of the same cultivars grown in different cities. To assess the extent to which our resource values agree with other estimates for the same species, we correlated our nectar resource data with values generated using the same methods recently published by Baude et al. [57].
PLOS ONE | DOI:10.1371/journal.pone.0158117 June 24, 2016
4 / 37
Quantifying Floral Resources in Urban Flower Meadows
(i) Nectar sampling. On days with no rain, nectar was allowed to accumulate for 24 hours in flowers from which insects were excluded by fine netting. Nectar quantification followed one of two protocols (see S2 File for detailed methods). When possible, nectar was collected directly using 1.0 μl microcapillaries (VWR International, UK) and the degrees Brix sugar content (g of sugar in 100g solution) estimated using a sucrose refractometer (VWR International, UK, using refractometers modified by the maker to accept very low volumes). Nectar sugar content of each sample (mg) was calculated from the equation s = 10dvC, where v is the volume of nectar (ml), and d is the density of a sucrose solution at a concentration C (g sucrose/100 g solution) as read on the refractometer [59,60]. The density was obtained as d = 0.0037921C + 0.0000178C2 + 0.9988603 [60]. We sampled 9.5 ± 0.6 (range 3–22) flowers per species (depending on availability) to estimate a species mean (see S1 Table). Where volumes of nectar were too small or concentrations of nectar too high to use this approach, nectar sugar was harvested by using a fine Gilson pipette to flush flowers with a known volume of distilled water and the Brix sugar content of the resulting solution estimated using a sucrose refractometer [57]. The mass of sugar contained was estimated as above. Nectar reward values are presented as mean sugar mass/floral unit ± 1 standard error of the mean. Nectar analyses covered 80% of species and over 99% of all flowers recorded in our surveys. (ii) Pollen sampling. Pollen per flower was estimated for all species using flowers collected in Edinburgh treatment meadows in 2012 and 2013, for a total of 15,925 flowers across 64 species (see S3 File for detailed methods). Pollen was harvested using sonication (Dawe sonicleaner) from flowers that opened and dehisced in the lab. The collected pollen was suspended in 70% ethanol, and the number of pollen grains in a known volume aliquot counted on a haemocytometer slide. We estimated the volume of a sample of 100 pollen grains of each species using the formula for an ellipsoid (Volume = (4/3) π (A/2) (B/2)2, where ‘a’ is the major axis and ‘b’ the minor axis of the pollen grain). Pollen volume per floral unit was then calculated as (total number of pollen grains/floral unit) x (mean volume per pollen grain). Sample size information for resource quantification in each species is provided in S1 Table, from a mean of 10.3 ± 0.6 (range 2–22) floral units/species, depending on availability. Pollen resource per floral unit per 24h was estimated by dividing the pollen volume per floral unit by floral longevity in days. This approach does not assume that pollen release is constant during the lifetime of each flower, but does assume that the proportion of flowers in the sampled population in each stage of pollen release (if it varies through floral life) was stable at the time longevity was measured (see below). Pollen resource values are presented as mean pollen volume/floral unit ± 1 standard error of the mean. Pollen resource per 24h was estimated for 78% of the species recorded in treatment plots, representing over 99% of the flowers present in our surveys (see S1 Table). We explored the predictability of pollen volume per flower (or per floret for Asteraceae) from floral morphology, using a linear model incorporating anther size and number of stamens as predictors in the base R package [61]. To simplify use of the model, we allocated anther sizes for each species to one of four size classes (S3 Table): ‘tiny’ (for all Asteraceae), ‘small’ (e.g. Galium verum, Stellaria media; to a maximum anther volume of ca. 1.25mm3), ‘medium’ (e.g. Ranunculus spp., Malcolmia maritima; ca. 1.25–2.25mm3) and ‘large’ (e.g. Papaver rhoeas, Chamerion angustifolium: volume > 2.25mm3). Data on stamen numbers/floral unit are also provided in S3 Table. Normalising transformations for the parameters (loge pollen per flower and 1/stamen number) were estimated by likelihood ratio tests using the powerTransform function in {car} [62], which maps variables to a family of functions to create a rank-preserving transformation of data using power functions. The regression model was fitted and checked for non-violation of assumptions of linearity, normality and homoscedasticity by inspection of residuals, fitted values and quantile plots.
PLOS ONE | DOI:10.1371/journal.pone.0158117 June 24, 2016
5 / 37
Quantifying Floral Resources in Urban Flower Meadows
We tested the predictive power of our model by estimating per-flower pollen volumes for a validation data set comprising 43 species quantified using the same lab protocol (Baude et al. unpublished data; see [57] for information on the Agriland Project generating these data). Species in the validation data set were selected to span the same range of measured pollen volumes. We attempted to develop similar predictive models for nectar sugar per flower based on the floral morphology traits of corolla diameter and corolla length. No useful predictive model resulted from this approach. (iii) Floral longevity. Floral unit longevity was estimated by scoring the numbers of newly opening and newly closed (i.e. post-reproductive) flowers entering a fixed survey area (see S3 File for detailed methodology). Under the assumption that the population of flowers of a given species is approximately stable, floral longevity in days can be estimated as (2a+(b-c)) / ((b-c) d), where a = total open flowers on first observation, b = total newly-open flowers on second observation, c = total newly-closed flowers on second observation, and d = number of days between first and second observation. To meet the assumptions of this approach as closely as possible, species were observed in the middle of their flowering period. To minimise weather impacts on resources, observations were set up when the weather forecast was for no rain for the following day and the temperature was above 15°C. Sampling included the full range of flower ages, totalling approximately 50 flowers distributed on at least five plants. After marking the boundary of the study area and/or the inflorescences to be included, the area was surveyed on two occasions exactly 24 hours apart (or a multiple thereof for longer-lasting flowers). Longevity data for almost all species were collected for plants from Edinburgh meadows, with additional data where necessary from sites specified in S1 Table, sampled using the same protocol.
Surveys of floral abundance and resources per meadow We used quadrat surveys to quantify floral abundance in all meadows, at the level of individual flowers for all species except Asteraceae, which were sampled at the level of the capitulum. To identify an appropriate sampling scheme for meadow-wide resources, we conducted pilot surveys of A2 meadows in their first year at three Edinburgh sites in 2012 (Inch, Cairntows, Davidson’s Mains). We sampled between 49 and 99 quadrats for each meadow, and then used computer simulations (sampling actual transect values at random, without replacement) to assess the impact of sampling increasing numbers of 1m2 quadrats on the cumulative mean of nectar sugar mass and pollen volume per meadow (S1 and S2 Figs). In each meadow, variance in mean estimates declined sharply with increasing sample size, stabilising for each meadow and resource at a sample size of ca. 20 quadrats. We therefore sampled 20 x 1m2 quadrats per treatment replicate per sampling interval for the Edinburgh meadows during the main 2013 survey season (a total of 1400 quadrats over all treatments and sampling intervals). The 20 quadrats were sampled at 4 m intervals along the length of the plot, alternating between edge quadrats (i.e. extending from the plot edge to 1m into the plot) and internal quadrats (i.e. extending from 1m to 2m into the plot). For the three other cities, time constraints restricted sampling to a subset of the 20-quadrat scheme, comprising seven edge quadrats per treatment replicate per sampling interval located at 8m intervals along the plot length. We compare the consequences of adopting the seven or 20-quadrat scheme below using the Edinburgh data. In short, we consider the seven-quadrat estimates adequate for comparison of mean differences between treatments, and all comparisons among cities are based on data for the seven-quadrat scheme. Total numbers of flowers (or capitula) per treatment replicate for each species were estimated as the product of mean floral density/m2 x meadow area (m2). In 2013, over the four cities, 80 plots and six sampling intervals, we recorded over two million flowers of 105 plant species, from a total of 3068 quadrats.
PLOS ONE | DOI:10.1371/journal.pone.0158117 June 24, 2016
6 / 37
Quantifying Floral Resources in Urban Flower Meadows
To illustrate variation among replicates we present detailed data summaries for each of the Edinburgh treatment replicates, and present the data for each meadow location in the other three cities in S4 and S5 Figs. To allow comparison between cities, treatments and sampling time points, we use means calculated for each treatment across all time points, replicates and cities, as appropriate. We excluded from these values data for four sites at which management issues (detailed below) led to complete failure of the treatment relative to others in the same group. These comprise two perennial sites and two A1 annual sites (S2 Table). Resources per meadow (nectar sugar mass or pollen volume) were calculated as the sum across plant species of (resource per flower) x (the number of flowers in the treatment replicate). 95% confidence intervals for resources are based on error estimates in the floral counts.
Statistical methods Statistical analyses were carried out in R version 3.1.3 (R Foundation for Statistical Computing, Vienna, Austria. 2015). To visualise seasonal trends in resource availability, means and 95% confidence intervals were plotted with the {ggplot2} package [63]. Correlations between each of nectar sugar and pollen volume and flower counts per transect were plotted with {PerformanceAnalytics} [64], and ranked dotplots were plotted with {stats} [65]. Flower counts (per species and summed across species) and resource values showed non-normality and heteroscedasticity that could not be corrected using standard data transformations. Variation of floral resource estimates among treatments was therefore tested using global Kruskal-Wallis tests on means calculated across all surveys for a given type (i.e. n = 7 quadrats x 5 replicates/city x 4 cities x 6 survey rounds) in {PMCMR} [66]. Pairwise comparisons between treatment means were tested using the post-hoc Tukey-Kramer-Nemenyi test with Chi-square correction for ties in {PMCMR}. Variation in the composition of planted floral meadows within a meadow treatment between cities and sampling points was tested using the manyglm function of the R package {mvabund} [67]. This approach allows identification of multivariate patterns and fitting of a separate Generalised Linear Model (GLM) for each flowering species using a common set of explanatory variables. Resampling-based hypothesis testing within mvabund was then used to make community-level and taxon-specific inferences about which factors were associated with the multivariate abundance patterns. For these analyses we used mean floral counts/m2 from seven quadrats for each treatment replicate (Annual A1, Annual A2, Perennial) per survey round (May, June, July, early August, late August, September) per city (Bristol, Edinburgh, Leeds, Reading). For this dataset we specified a negative binomial error distribution, and checked assumptions of mean-variance and log-linearity as detailed in [67], both by plotting directly and by plotting residuals versus fitted values. We used Monte Carlo bootstrapping to estimate p-values adjusted to control the family-wise error rate across species, at the default setting of 1000 resamples. Variation in floral composition was visualised using non-metric multidimensional scaling (NMDS) with the metaMDS function of {vegan} [68], using a Bray-Curtis distance matrix calculated using mean floral counts (capitula for Asteraceae) per species per m2 for each meadow location and sampling time point. Meadows with fewer than 10 flowers (or capitula) in total per m2 were excluded from the analysis. Because NMDS only uses rank information and maps ranks non-linearly onto ordination space, it can handle non-linear species responses of any shape and robustly find underlying gradients. As the final ordination is partly dependent on the initial configuration, we used up to 99 consecutive NMDS iterations with random starting configurations to test for stability of the result. All raw data, metadata and R code are available from the authors on request.
PLOS ONE | DOI:10.1371/journal.pone.0158117 June 24, 2016
7 / 37
Quantifying Floral Resources in Urban Flower Meadows
Results Resource per flower (a) Nectar sugar. Nectar sugar estimates per 24h for the seed mix species and associated weeds are shown in Fig 1 and S1 Table. Most of the highest values were for Asteraceae, for which a floral unit comprises an inflorescence (capitulum) rather than a single flower (Fig 1). The top-ranked annual species/floral unit/24h were Centaurea cyanus (896 ± 174μg s.e.m.), Cosmos bipinnatus (701 ± 69μg) and Calendula officinalis (470 ± 11μg). The top-ranked perennial species were Leontodon hispidus (1827 ± 193μg), Centaurea nigra (1474 ± 76μg) and Echium vulgare (688 ± 103μg). The top-ranked weed species, all native Asteraceae, produced more nectar sugar per floral unit than any seed mix species: Senecio jacobaea (2921 ± 448μg), Cirsium arvense (2609 +/- 239μg), Cirsium vulgare (2323 ± 418μg), and Taraxacum agg. (2137 ± 286μg). Both the annual and perennial mixes contained species with very low nectar sugar rewards per floral unit, including species with large individual flowers such as Papaver rhoeas (0.6 ± 0.6μg) and Eschscholzia californica (10 ± 1μg) in the annual mix, and species sampled as small individual flowers such as Lobularia maritima (4 ± 1μg) in the annual mix and Daucus carota (27 ± 7μg) and Galium verum (3.2 ± 0.6μg) in the perennial mix.
Fig 1. Mean nectar sugar mass per 24h per floral unit for species in A. the annual seed mix, B. the perennial seed mix, and B. native weeds in either mix. Values shown are ranked means in each group (mean values and standard errors are provided in S1 Table). Images of the top ranked species in each group are shown, with the highest-ranked at right. Images are provided by the project team with the exceptions of Echium vulgare (author: Ewan Cole) provided in 2016 under a (CCAL) CC BY 4.0 license from the Urban Flora of Scotland project. doi:10.1371/journal.pone.0158117.g001
PLOS ONE | DOI:10.1371/journal.pone.0158117 June 24, 2016
8 / 37
Quantifying Floral Resources in Urban Flower Meadows
(b) Pollen. Total pollen rewards per floral unit were highest in Asteraceae sampled at the level of the entire capitulum (Leucanthemum vulgare, 15.9 ± 2μl, Cosmos bipinnatus 13.8 ± 1.9μl), and lowest in those species sampled at the level of small individual flowers (e.g. Myosotis arvensis, 0.0004 ± 0.0001μl) (S1 Table). Of species sampled at the level of a single flower, by far the most rewarding were the poppies, Papaver rhoeas (13.3 ± 2.8μl) and Eschscholzia californica (8.3±1.1μl). Among weed species, values were again highest for Asteraceae, including Glebionis segetum (5.1 ± 0.9μl) (presumed to be a garden escape) and native Taraxacum agg. (2.8 ± 0.7μl). Quantifying the contribution of each species to daily pollen resource provision at the meadow level requires scaling of total pollen volume by floral longevity, which ranged from mean values of a single day (e.g. Cerastium fontanum, Veronica persica, Vicia hirsuta) to 14.8 days (Leucanthemum vulgare), and was generally higher in Asteraceae species sampled at the level of the capitulum (see Fig 2 and S1 Table). The highest pollen rewards/floral unit/24h were provided by annual mix species; flowers of Papaver rhoeas (6.0μl pollen) provided more than twice the values for the next-ranked species, Eschscholzia californica (2.4μl) and Calendula officinalis (1.8μl; this species ranked highly for both nectar and pollen rewards). In the annual mix, Coreopsis picta (0.7μl) had a very low pollen reward per capitulum in comparison to other
Fig 2. Mean pollen volume per 24h per floral unit for species in A. the annual seed mix, B. the perennial seed mix, and C. native weeds in either mix. Values shown are ranked means in each group. Mean values and standard errors, longevity and pollen volume/floral unit are provided in S1 Table). Images of the top ranked species in each group are shown, with the highest-ranked at right. Images are provided by the project team with the exceptions of Chamerion angustifolium (author: Ewan Cole) provided in 2016 under a (CCAL) CC BY 4.0 license by the Urban Flora of Scotland project. doi:10.1371/journal.pone.0158117.g002
PLOS ONE | DOI:10.1371/journal.pone.0158117 June 24, 2016
9 / 37
Quantifying Floral Resources in Urban Flower Meadows
Asteraceae sampled. The top-ranked perennial mix species by floral unit were Malva moschata (2.3μl), Centaurea nigra (2.1μl; this species ranked highly for both nectar and pollen rewards) and Leucanthemum vulgare (1.1μl). The top-ranked weed species in our meadows were native Taraxacum agg. (1.25μl) and Chamerion angustifolium (0.7μl). (c) Statistical prediction of resource per flower from floral morphology. Laboratory estimates of total pollen volume per floral unit were predicted by a linear model using anther size and stamen number as predictors (S3 Fig; adjusted R2 = 0.7025, F4,54 = 35.23, p
